1 |
On Wednesday 22 September 2010 02:24:39 Grant wrote: |
2 |
> >> >>>> I'm getting a lot of machine check exception errors in dmesg on my |
3 |
> >> >>>> hosted server. Running mcelog I get: |
4 |
> >> >>>> ... |
5 |
> >> > |
6 |
> >> > They offered to take my machine down and do a memory test which they |
7 |
> >> > said would take a number of hours. Is a memory test likely to help? |
8 |
> >> > Did you suggest reseating or replacing RAM modules as opposed to a |
9 |
> >> > memory test because it will result in less downtime? |
10 |
> >> |
11 |
> >> I suspect that your hosting provider are offering you this memory test |
12 |
> >> because they don't want to go swapping out memory modules willy-nilly. |
13 |
> >> |
14 |
> >> How do they know that the problem is really memory, and not your |
15 |
> >> operating system? If they take all this RAM out and put new RAM in, |
16 |
> >> what do they do with the old RAM? They don't know if it's good or bad, |
17 |
> >> so are they expected to just slap it in a server belonging to another |
18 |
> >> customer, and stitch him up? |
19 |
> >> |
20 |
> >> A memory test is likely to identify bad RAM, if it is bad, so you should |
21 |
> >> proceed with this. This is likely the best route to solving the problem. |
22 |
> >> |
23 |
> >> I think that ideally, for you, they would move the system image onto a |
24 |
> >> different known-good server with the same configuration. Then you cannot |
25 |
> >> complain if the same problems start occurring again. If the problem is |
26 |
> >> genuinely hardware then they won't. And the hosting provider is free to |
27 |
> >> run diagnostics on your old machine. |
28 |
> >> |
29 |
> >> But realistically, the memory test is likely to show up a bad RAM |
30 |
> >> module, you'll get it replaced and be up and running within a few |
31 |
> >> hours. Why would you refuse? If your system needed a guaranteed uptime |
32 |
> >> you'd perhaps have to pay for a higher level of service than the fees |
33 |
> >> you're paying at present. |
34 |
> > |
35 |
> > I run memory tests overnight. If a module is seriously borked then it |
36 |
> > will fail earlier. Reseating/replacing takes a few minutes, instead of |
37 |
> > hours. |
38 |
> > |
39 |
> > If they have spare machines (for dev't or testing) they can fit the |
40 |
> > memory module(s) there and test them exhaustively, before they put the |
41 |
> > good ones back into a customer's machine. |
42 |
> |
43 |
> Thanks Mick and Stroller. I'll see if they'll go for this. |
44 |
|
45 |
You're welcome. Bear in mind though that a lot of hosters are just glorified |
46 |
resellers with an account in a bigger data centre. In many cases they do not |
47 |
even have physical access to the machines. Only the data centre techies do |
48 |
and they may be less willing to oblige and break procedure or routine, just |
49 |
because one end user out of hundreds/thousands complained about some memory |
50 |
errors. |
51 |
|
52 |
YMMV |
53 |
-- |
54 |
Regards, |
55 |
Mick |